An Adaptive Typing Biometric System with Varying users Model

Abstract

Keystroke dynamics is a behavioral biometric parameter that can be fully exploit in order to build a continuous biometric authentication system that must comply with strict constraints in the use of computational resources, as well as energy, without lowering the security required level. This paper introduces a typing biometric system that continuously adjourns the users models for taking into account both short term amd long term modification in their habits. The system relies on a reduced space features that mainly uses the hold time and both the keyUp-keyDown and keyDown-keyUp time for some selected keys. An Adaptive Continuous Biometric Authentication Scheme, recomputes each user model according to his/her most recent typing history in a temporal sliding window of fixed dimension. Contrary to the most recent work in the literature, we do not limit to adapt the model of the user under authentication but we consider the potential evolution of the model of the other enrolled users, forecasting their possible evolution, again from the data in the sliding window. System performances have been tested with different window size and under different distance metric (namely Euclidean distance, Manhattan distance and cosine similarity). Moreover, it has been considered the balancing among used computational resource and accuracy. Experimentations on the usability of this approach are also reported.